- Email: [email protected]
Identification of personalized dysregulated pathways in hepatocellular carcinoma
G Model PRP-51730; No. of Pages 6
ARTICLE IN PRESS Pathology – Research and Practice xxx (2017) xxx–xxx
Contents lists available at ScienceDirect
Pathology – Research and Practice journal homepage: www.elsevier.com/locate/prp
Original article
Identification of personalized dysregulated pathways in hepatocellular carcinoma Hong Li 1 , Xiumei Jiang 1 , Shengjie Zhu, Lihong Sui ∗ Department of Oncology, Weihai Central Hospital, Weihai, 264400, Shandong, PR China
a r t i c l e
i n f o
Article history: Received 6 May 2016 Keywords: Hepatocellular carcinoma Differentially expressed genes Personalized dysregulated pathways Normal distribution analysis
a b s t r a c t Introduction: Hepatocellular carcinoma (HCC) is the most common liver malignancy, and ranks the fifth most prevalent malignant tumors worldwide. In general, HCC are detected until the disease is at an advanced stage and may miss the best chance for treatment. Thus, elucidating the molecular mechanisms is critical to clinical diagnosis and treatment for HCC. The purpose of this study was to identify dysregulated pathways of great potential functional relevance in the progression of HCC. Materials and methods: Microarray data of 72 pairs of tumor and matched non-tumor surrounding tissues of HCC were transformed to gene expression data. Differentially expressed genes (DEG) between patients and normal controls were identified using Linear Models for Microarray Analysis. Personalized dysregulated pathways were identified using individualized pathway aberrance score module. Results: 169 differentially expressed genes (DEG) were obtained with |logFC| ≥ 1.5 and P ≤ 0.01. 749 dysregulated pathways were obtained with P ≤ 0.01 in pathway statistics, and there were 93 DEG overlapped in the dysregulated pathways. After performing normal distribution analysis, 302 pathways with the aberrance probability ≥ 0.5 were identified. By ranking pathway with aberrance probability, the top 20 pathways were obtained. Only three DEGs (TUBA1C, TPR, CDC20) were involved in the top 20 pathways. Conclusion: These personalized dysregulated pathways and overlapped genes may give new insights into the underlying biological mechanisms in the progression of HCC. Particular attention can be focused on them for further research. © 2017 Elsevier GmbH. All rights reserved.
1. Introduction Hepatocellular carcinoma (HCC) is the most common liver malignancy, and ranks the fifth most prevalent malignant tumors worldwide [1]. Despite recent progress in anticancer, only 7% of patients with advanced HCC can expect to live for 5 years. In general, HCC are detected until the disease is at an advanced stage and may miss the best chance for treatment. Therefore, elucidating the molecular mechanisms is critical to clinical diagnosis and treatment for HCC. It is generally accepted that the occurrence of HCC is the result of genomic alteration [2,3]. In recent years, microarray technology has been used as an advanced high-throughput strategy for detecting gene expression profiles [4]. Tackels-Horne and colleagues successfully identified 842 over-regulated genes and 343 down-regulated
∗ Corresponding author at: Weihai Central Hospital, No.3, Mishan West Road, Wendeng District, Weihai 264400, Shandong, PR China. E-mail address: [email protected] (L. Sui). 1 These authors contributed equally.
genes involved in the pathogenesis of HCC [5]. Similarly, Chen and co-workers made a cross-study comparison for the tumor tissues and non-tumor tissues, and the results indicated that 1640 genes were significantly differentially expressed in the tumor tissues [6]. In addition, several growth factors, including EGFR, HDGF and IGF, have been proposed to be involved in the progression of HCC [7–9]. A previous study has demonstrated that FOXO-TXNIP pathway played a pivotal role in the inhibition of HCC growth by MK-801 [10]. Previous studies have proved that many critical genes and pathways are dysregulated during cancer initiation and progression [11]. Although there have been numerous studies on HCC pathogenesis, the results are not uniform and share only a small number of potential genes and pathways. The shortcoming highlights the importance of personalized pathway analysis. However, most current pathway analyses are mainly focused on discovering dysregulated pathways between two phenotype groups. Based on the shortcoming, a new approach is proposed to identify dysregulated pathways in an individual case. Our proposed method is based on the comparison of one cancer sample with many accumulated normal samples (we use “nRef” to refer to the accumulated normal
http://dx.doi.org/10.1016/j.prp.2017.01.015 0344-0338/© 2017 Elsevier GmbH. All rights reserved.
Please cite this article in press as: H. Li, et al., Identification of personalized dysregulated pathways in hepatocellular carcinoma, Pathol. – Res. Pract (2017), http://dx.doi.org/10.1016/j.prp.2017.01.015
G Model PRP-51730; No. of Pages 6
ARTICLE IN PRESS
2
H. Li et al. / Pathology – Research and Practice xxx (2017) xxx–xxx
samples). Personalized identification of dysregulated pathways is an important step toward understanding the disease mechanisms. An important clinical aspect of our method is providing a pathway interpretation of a single cancer, even though matched normal data are unavailable. Finally, individualized dysregulated pathways were identified by quantifying the aberrance of an individual sample’s pathway by comparing it with accumulated normal samples. In this study, differentially expressed genes (DEG) and individualized dysregulated pathways were identified. Microarray data of HCC patients were downloaded from ArrayExpress database [12]. The data were then preprocessed, and DEGs were identified using the LIMMA package. Subsequently, significantly dysregulated pathways between normal and cancer groups were identified. Finally, personalized dysregulated pathways were analyzed by using accumulated normal data. Among the 72 patients with HCC, the aberrance probability of each pathway was calculated. This study might provide useful information for exploring critical genes and individualized altered pathways, which can give insights into the diagnosis and treatment of HCC.
2.3. Pathway analysis Pathway analysis has become the first choice for capturing biologically and clinically relevant information. In our study, an individualized pathway aberrance score (iPAS) model was proposed to identify dysregulated pathways [21]. It makes special use of accumulated normal data to serve as the nRef. Data-processing procedures were described in more detail as follows. 2.3.1. Gene-level statistics The microarray data of normal samples were normalized using the quantile normalization method in preprocessCore package. For normal controls, the mean and standard deviations of the gene expression level were calculated. For individual tumor cases, quantile normalization [15] was performed after combining the single tumor microarray with all nRef samples. The gene-level statistics of individual tumor samples were standardized using the mean and standard deviation of the reference. The formula is shown as the following: zi =
2. Materials and methods 2.1. Datasets 2.1.1. Gene expression data and preprocessing Microarray expression data were downloaded from ArrayExpress database (http://www.ebi.ac.uk/arrayexpress/) using accession number E-GEOD-39791 [13]. The dataset included gene expression data from tumor and matched non-tumor surrounding tissues of 72 HCC patients who underwent surgical resection as the primary treatment. Gene expression data was generated using the HumanHT-12 Version 4.0 Expression Beadchip (Illumina) according to the annotation files. Linear Models for Microarray Data (LIMMA) was chosen to reprocess gene expression data. To eliminate influences of nonspecific hybridization, the background was corrected using robust multichip average (RMA) [14]. Normalization was performed with quantiles function. Then we used the MAS method to correct perfect match (PM)/mismatch (MM) [15], and the Medianpolish function was used to summarize expression data [16]. Then probes were mapped to gene symbols. The probe was discarded if it could not match any genes. The levels of probes were averaged as the final gene expression value if more than one probe was mapped to a single gene [17].
2.1.2. Pathway data All Homo sapiens-related biological pathway data were downloaded from Reactome database (http://www.reactome.org/) [18]. In general, pathways with a large number of genes are not easily understood by human experts [19]. Therefore, the pathways with a gene size >100 were filtered out. Meanwhile, the pathways without transformational gene symbols were removed. As a result, a total of 1022 pathways, including 4928 genes, were obtained for further analysis.
2.2. DEG analysis In the present study, Linear Models for Microarray Analysis (LIMMA) package of R were used to identify DEG between HCC patients and normal controls [20]. The values of |logFC| ≥ 1.5 and P ≤ 0.01 were selected as the cut-off criteria. In addition, the DEG belonging to pathways were ascertained by taking the intersection of identified DEG and all genes in pathways.
gTi − mean (gnRef ) stdev (gnRef )
(1)
where gTi represents the expression value of i-th gene belonging to the tumor cases, where mean (gnRef ) and stdev (gnRef ) represent the mean and standard deviation of the gene expression value of all reference samples, respectively. 2.3.2. Pathway-level statistics For each specific pathway, the standardized gene-level statistics of all genes were extracted. Then the gene-level statistics for all genes in a pathway were aggregated into a single pathway-level statistic. The model is as follows: iPAS =
zin
(2)
where zi symbolizes the standardized expression value of i-th gene belonging to the pathway, where n symbolizes the number of genes belonging to the pathway. 2.3.3. Dysregulated pathways Significance test was performed to assess dysregulated pathways associated with HCC. A generalized two-sample wilcoxon test was applied to assess the statistical significance of the pathwaylevel statistic. The false discovery rate (FDR) was used to correct the significance level [22,23]. The pathway with P < 0.01 was considered to be significantly dysregulated between normal and cancer groups. Moreover, a hierarchical clustering algorithm was applied to grouping the dysregulated pathways and the results were visualized with TreeView [24]. In addition, we obtained the common genes between known disease genes and dysregulated pathway genes. The known genes related to liver disease were selected, and a total of 39 genes were downloaded [25]. The pathways, including disease genes, were also ascertained. 2.3.4. Individualized dysregulated pathways Significance can be obtained against null distribution generated from normal samples on the basis of pathway-level statistics. The statistic of the null distribution was acquired via comparing every normal sample with all nRef samples. Then the p-value of each pathway was obtained after comparing a single cancer case with the null distribution generated above. A value with P < 0.05 was considered to indicate a statistically significant difference. Among the 72 patients with HCC, the aberrance probability of each significant pathway (P < 0.05) was calculated and ranked. The pathways with higher aberrance probability were considered to be more prone to alterations.
Please cite this article in press as: H. Li, et al., Identification of personalized dysregulated pathways in hepatocellular carcinoma, Pathol. – Res. Pract (2017), http://dx.doi.org/10.1016/j.prp.2017.01.015
G Model
ARTICLE IN PRESS
PRP-51730; No. of Pages 6
H. Li et al. / Pathology – Research and Practice xxx (2017) xxx–xxx
3
Table 1 Dysregulated pathways and the distribution of DEG. Term
Number of DEG
P-value
Probability
Phase 1 − Functionalization of compounds Phase II conjugation Cytochrome P450 − arranged by substrate type Xenobiotics Complement cascade Bile acid and bile salt metabolism CYP2E1 reactions Glucose metabolism Formation of Fibrin Clot (Clotting Cascade) Lipoprotein metabolism Regulation of Complement cascade Terminal pathway of complement
13 11 10 7 5 5 5 4 4 4 4 4
3.38E-10 1.55E-08 2.73E-10 7.49E-11 1.89E-10 3.48E-08 5.49E-11 6.62E-03 2.92E-10 6.11E-11 1.37E-08 3.62E-10
0.014 0.111 0.014 0 0.083 0.069 0 0.097 0.042 0 0.083 0.056
DEG denotes differential expressed genes.
Table 2 Known genes related to liver disease as well as shared in dysregulated pathways. No.
Genes
No.
Genes
No.
1 2 3 4 5 6 7 8 9 10 11 12
CCND1 CDKN1A PCNA IFNG IL6 ALB MET AXIN1 CASP3 SLC17A5 BCL2 JUN
13 14 15 16 17 18 19 20 21 22 23 24
CDH1 CTNNB1 PLAU FAS PTGS2 TP53 APC TGFB1 FASLG VEGFA TNF PIK3CA
25 26 27 28 29 30 31 32 33 34 35
3. Results 3.1. DEG analysis According to the criteria outlined (|logFC| ≥ 1.5, P ≤ 0.01), a total of 169 DEG were identified between tumor cases and normal controls, of which 17 were up-regulated and 152 were down-regulated. Among the DEG, 93 genes were as well as shared in pathways.
are shown in Table 3. Among them, degradation of the extracellular matrix and apoptotic cleavage of cellular proteins pathway contain 5 and 4 disease genes, respectively. 3.2.2. Individual dysregulated pathways Significance was obtained by null distribution calculated from the nRef. Statistical data from a single tumor sample were compared with the null distribution to yield p-value. Thus, the p-value of each pathway was obtained, and the pathway with p < 0.05 was screened. Subsequently, the aberrance probability of each significant pathway (p < 0.05) among the 72 patients was calculated and ranked. By ranking the pathways according to the aberrance probability, top 20 pathways are listed in Table 4. Based on our results, the “protein folding” pathway showed the highest aberrance probability (0.931). The other three pathways with the aberrance probability of 0.917 were also more prone to dysregulation, which were RNA Polymerase II Pre-transcription Events, Viral Messenger RNA Synthesis and Transport of Ribonucleoproteins into the Host Nucleus. Among the 72 patients, 933 pathways are able to dysregulate. A total of 302 pathways are more prone to alterations, the aberrance of which is more than 0.5. For further analysis, it was found that the top 20 ranked pathways were only involved in three DEG (TUBA1C, TPR, CDC20), and the other pathways do not contain DEG.
3.2. Pathway analysis 4. Discussion 3.2.1. Dysregulated pathways Wilcoxon test was performed to identify dysregulated pathways between normal controls and tumor cases. After FDR correction, a total of 749 pathways were significant (P < 0.01), of which only 165 pathways contain DEG and the remaining 584 pathways do not contain DEG. These pathways were ranked according to the number of DEG, and top 12 ranked pathways are listed in Table 1. It was also found that the aberrance probability of top ranked pathways was less than 0.12. For instance, the Phase 1 − Functionalization of compounds pathway, including 13 DEG, but the aberrance of which is only 0.014. Besides, a hierarchical clustering algorithm was applied to grouping the pathways and samples. The results for the 749 dysregulated pathways are shown in Fig. 1, and two features are evident. First, with a few exceptions, the clinical samples can be divided into two major clusters, one representing tumor samples, and the other representing normal samples. Second, pathways varied significantly among the tumor and normal samples. Furthermore, we obtained 39 disease genes, of which 35 genes (Table 2) were also shared in dysregulated pathways, such as CCND1, CDKN1A and PCNA. Meanwhile, the pathways containing disease genes were also ascertained. The results showed that there were 169 pathways including disease genes. The pathways were ranked by the number of disease genes, and top 9 ranked pathways
Cancers are generally caused by mutation of multiple genes or dysregulation of pathways. Identifying underlying genes and pathways would help to understand and diagnose cancers. The accumulation of microarray data makes it possible to detect critical genes and pathways in a more efficient way. This study aimed at analyzing underlying genes and personalized dysregulated pathways involved in HCC. LIMMA package was used to identify DEG between HCC patients and normal controls. The iPASbased approach was applied to explore individualized dysregulated pathways. In summary, many underlying genes and pathways were identified in our research. As a new approach for pathway analysis, iPAS was developed to identify altered pathways in an individual by making using of accumulated normal samples. Our proposed method captures biologically and clinically relevant information in a more accurate way compared to the interpretation of a single cancer sample in the context of cancer-only cohort. It is also useful in classifying unknown samples into cancer or normal groups. This study provides useful information to explore the dysregulated pathways and the underlying molecular mechanisms of HCC. Based on our results, a total of 169 genes were identified under the criteria of |logFC| ≥ 1.5 and P ≤ 0.01, of which 17 were up-regulated and 152 were down-regulated. Among the DEG,
Please cite this article in press as: H. Li, et al., Identification of personalized dysregulated pathways in hepatocellular carcinoma, Pathol. – Res. Pract (2017), http://dx.doi.org/10.1016/j.prp.2017.01.015
G Model PRP-51730; No. of Pages 6
ARTICLE IN PRESS
4
H. Li et al. / Pathology – Research and Practice xxx (2017) xxx–xxx
Fig 1. A hierarchical cluster for the 749 deregulated pathways. Rows represent individual pathways and columns represent individual tissue samples. The color reflects different samples, red representing normal samples, and blue representing tumor samples.
Table 3 Dysregulated pathways and the distribution of disease genes. Term
Disease genes
Probability
Degradation of the extracellular matrix Apoptotic cleavage of cellular proteins Degradation of beta-catenin by the destruction complex Transcriptional regulation of white adipocyte differentiation Cyclin D associated events in G1 G1 Phase Intrinsic Pathway for Apoptosis Death Receptor Signalling Beta-catenin phosphorylation cascade
CASP3,CDH1,MMP2,MMP9,TIMP1 APC,CASP3,CDH1,CTNNB1 APC,AXIN1,CTNNB1 PPARG,TGFB1,TNF CCND1,CDKN1A,CDKN2A CCND1,CDKN1A,CDKN2A BCL2,CASP3,TP53 FAS,FASLG,TNF APC,AXIN1,CTNNB1
0.292 0.292 0.861 0.708 0.806 0.806 0.806 0.306 0.375
Table 4 The top 20 ranked pathways and distribution of DEG. Pathways
No.
Probability
DEG
Protein folding RNA Polymerase II Pre-transcription Events Viral Messenger RNA Synthesis Transport of Ribonucleoproteins into the Host Nucleus Transcription of the HIV genome HIV Transcription Initiation RNA Polymerase II HIV Promoter Escape RNA Polymerase II Promoter Escape RNA Polymerase II Transcription Initiation RNA Polymerase II Transcription Initiation And Promoter Clearance RNA PolymeraseII Transcription Pre-Initiation And Promoter Opening Transport of the SLBP Dependant Mature mRNA Transport of the SLBP independent Mature mRNA APC/C:Cdc20 mediated degradation of Cyclin B Translesion Synthesis by POLH APC/C-mediated degradation of cell cycle proteins Regulation of mitotic cell cycle
67 66 66 66 65 65 65 65 65 65 65 65 65 65 65 64 64
0.931 0.917 0.917 0.917 0.903 0.903 0.903 0.903 0.903 0.903 0.903 0.903 0.903 0.903 0.903 0.889 0.889
TUBA1C – TPR TPR – – – – – – – TPR TPR CDC20 – CDC20 CDC20
93 were also shared in pathways. Subsequently, pathway analysis was carried out to assess potential pathways associated with HCC. Wilcoxon test was applied to identify significantly dysregulated pathways between normal controls and tumor cases. The results indicated that 749 pathways with p < 0.01 were significantly altered. Among these pathways, only 165 pathways contain DEG.
It was found that the number of DEG is inconsistent with statistical significance. It is possible that genes do not work in isolation but interact with each other to perform their functions. Therefore, dysregulated pathways can serve as better biomarkers compared with single genes.
Please cite this article in press as: H. Li, et al., Identification of personalized dysregulated pathways in hepatocellular carcinoma, Pathol. – Res. Pract (2017), http://dx.doi.org/10.1016/j.prp.2017.01.015
G Model PRP-51730; No. of Pages 6
ARTICLE IN PRESS H. Li et al. / Pathology – Research and Practice xxx (2017) xxx–xxx
In recent years, many genes have been identified to be related to liver disease. In this article, 39 disease genes were downloaded, of which 35 genes were also shared in dysregulated pathways. These genes once again were validated, such as CCND1, CDKN1A and PCNA. Meanwhile, the pathways containing disease genes were also ascertained, and only 169 dysregulated pathways include disease genes, such as degradation of the extracellular matrix and apoptotic cleavage of cellular proteins pathway. This indicates that some key genes cannot be discovered, and more attention can be paid to them. These potential genes and pathways may give insights into disease mechanisms and even drug targets. CCND1 is a protein coding gene and encodes the cyclin-D1 protein, which is responsible for cell cycle progression [26]. In G1 phase, cyclin-D1 is synthesized rapidly by the inducing action of mitogenic growth factors and accumulated in the nucleus[27]. Experimental study revealed that Cyclin-D1 may play an important role in HCC [28]. In addition, amplification and over-expression of this gene play pivotal roles in the development of various tumors, such as non-small cell lung cancer [29], breast cancer [30], and prostate cancer[31]. CDKN1A (P21) is a cyclin-dependent kinase inhibitor, and can regulate cell cycle progression and cellular senescence [32,33]. The gene is an important intermediate by which p53 mediates its role as an inhibitor of cellular proliferation in response to DNA damage [34]. A previous study indicated that P21 was upregulated in metastatic canine mammary tumor [35]. P21 plays an important role during HCC progression [36]. PCNA is closely related to cell DNA synthesis and can be used as evaluation index for the cell proliferation state. The protein can interact with CDKN1A and plays a regulatory role in DNA replication and DNA damage repair [37]. The research investigating the relationship between PCNA and tumor has been a hot field, and PCNA is a potential therapeutic target in cancer therapy [38]. These known disease genes, which were also included in dysregulated pathways, can be considered as critical therapeutic target. Pathway-based identification of cancer has been proposed to be a robust method, especially individualized pathway analysis provides a more effective way in cancer identification. Null distribution was applied to yield the P −value of each pathway. The pathway with P < 0.05 was screened among the 72 tumor samples. Furthermore, the significant aberrance probability of each pathway was calculated and ranked. The results indicated that protein folding pathway had the highest aberrance probability. Protein folding pathway consists of 52 genes, of which TUBA1C is differentially expressed. Protein folding is an essential process in which polypeptide can self-assemble into its characteristic and functional three-dimensional structure. The correct three-dimensional structure is crucial to maintaining the protein function. Recent evidence suggests that misfolded protein can cause a variety of diseases, such as Phenylketonuria, Alzheimer and Parkinson’s diseases [39–41]. Pathways that are commonly dysregulated across the overwhelming majority of cancer patients may be critical in identifying cancer. These identified pathways were considered closely related to HCC. For further analysis, it was revealed that top 20 ranked pathways are only involved in three DEG (TUBA1C, TPR, CDC20). This is not consistent with traditional methods that detect DEG between tumor cases and normal controls. The power of pathway-level statistics not only depends on the proportion of DEG in a pathway, but is also involved in the size of the pathway and the amount of correlation between genes in the pathway. It is believed that dysregulated pathways show greater advantage than single genes in identifying cancer. CDC20 encodes cell-division cycle protein 20, which is a regulatory protein of cell division [42,43]. Its most important function is to activate the anaphase-promoting complex (APC/C), and plays an important role in tumorigenesis and progression of multiple tumors [44]. Accumulating evidence indicated that CDC20 may function as an oncogene and was over-expressed in various types
5
of human tumors, such as breast cancer, ovarian cancer and gastric cancer [45–47]. Increased expression of CDC20 has also been demonstrated to be associated with the development and progression of HCC [48]. However, TPR and TUBA1C show poor research on disease mechanisms. The three DEG (TUBA1C, TPR, CDC20), which also belong to the top ranked pathways, are considered pivotal in the progression of HCC. More attention should be paid to them, and they may be regarded as a promising therapeutic target for HCC. In this study, many underlying genes and personalized dysregulated pathways were identified. Among them, a part of them have been reported to be associated with HCC, and the others showed poor research. The results need to be validated in a prospective study, and all of these would become another perspective research. 5. Conclusions In this study, underlying genes and personalized dysregulated pathways were identified. First, LIMMA package was used to identify DEG and 169 genes were significantly differentially expressed. Next, the Wilcoxon test was performed to identify dysregulated pathways between normal and tumor groups. The analysis revealed that 749 pathways were significantly dysregulated between normal and cancer groups. Meanwhile, disease genes related to liver were also obtained. A total of 39 disease genes were downloaded, of which 35 were included in dysregulated pathways. Dysregulated pathways, in particular those pathways including disease genes, might provide insights into disease mechanisms and drug targets. Finally, an iPAS-based approach was employed to identify personalized dysregulated pathways. Individualized pathway analysis indicated that protein folding pathway had the highest aberrance probability. Furthermore, it was also revealed that the top ranked pathways were only involved in three DEG: TPR, TUBA1C and CDC20. Dysregulated pathways act as more effective biomarkers in identifying cancers. These personalized dysregulated pathways and overlap genes may give new insights into the underlying biological mechanisms driving the progression of HCC. Conflict of interest None. References [1] A. Jemal, F. Bray, M.M. Center, J. Ferlay, E. Ward, D. Forman, Global cancer statistics, CA. Cancer J. Clin. 61 (2011) 69–90. [2] X.Y. Guan, Y. Fang, J. Sham, D. Kwong, Y. Zhang, Q. Liang, et al., Recurrent chromosome alterations in hepatocellular carcinoma detected by comparative genomic hybridization, Genes Chromosomes Cancer 30 (2001) 110. [3] S.S. Thorgeirsson, J.W. Grisham, Molecular pathogenesis of human hepatocellular carcinoma, Nat. Genet. 31 (2002) 339–346. [4] A. Suriawinata, R. Xu, An update on the molecular genetics of hepatocellular carcinoma, Semin. Liver Dis. 24 (2004) 77–88. [5] D. Tackels-Horne, M.D. Goodman, A.J. Williams, D.J. Wilson, T. Eskandari, L.M. Vogt, et al., Identification of differentially expressed genes in hepatocellular carcinoma and metastatic liver tumors by oligonucleotide expression profiling, Cancer 92 (2001) 395–405. [6] X. Chen, S.T. Cheung, S. So, S.T. Fan, C. Barry, J. Higgins, et al., Gene expression patterns in human liver cancers, Mol. Biol. Cell 13 (2002) 1929–1939. [7] M.B. Thomas, D. Jaffe, M.M. Choti, J. Belghiti, S. Curley, Y. Fong, et al., Hepatocellular carcinoma: consensus recommendations of the national cancer institute clinical trials planning meeting, J. Clin. Oncol. 28 (2010) 3994–4005. [8] H. Enomoto, H. Nakamura, W. Liu, S. Nishiguchi, Hepatoma-Derived growth factor: its possible involvement in the progression of hepatocellular carcinoma, Int. J. Mol. Sci. 16 (2015) 14086–14097. [9] M. Enguita-German, P. Fortes, Targeting the insulin-like growth factor pathway in hepatocellular carcinoma, World J. Hepatol. 6 (2014) 716–737. [10] F. Yamaguchi, Y. Hirata, H. Akram, K. Kamitori, Y. Dong, L. Sui, et al., FOXO/TXNIP pathway is involved in the suppression of hepatocellular carcinoma growth by glutamate antagonist MK-801, BMC Cancer 13 (2013) 468.
Please cite this article in press as: H. Li, et al., Identification of personalized dysregulated pathways in hepatocellular carcinoma, Pathol. – Res. Pract (2017), http://dx.doi.org/10.1016/j.prp.2017.01.015
G Model PRP-51730; No. of Pages 6
ARTICLE IN PRESS
6
H. Li et al. / Pathology – Research and Practice xxx (2017) xxx–xxx
[11] A.H. Bild, G. Yao, J.T. Chang, Q. Wang, A. Potti, D. Chasse, et al., Oncogenic pathway signatures in human cancers as a guide to targeted therapies, Nature 439 (2006) 353–357. [12] H. Parkinson, M. Kapushesky, M. Shojatalab, N. Abeygunawardena, R. Coulson, A. Farne, et al., ArrayExpress–a public database of microarray experiments and gene expression profiles, Nucleic Acids Res. 35 (2007) D747–D750. [13] J.H. Kim, B.H. Sohn, H.S. Lee, S.B. Kim, J.E. Yoo, Y.Y. Park, et al., Genomic predictors for recurrence patterns of hepatocellular carcinoma: model derivation and validation, PLoS Med. 11 (2014) e1001770. [14] L. Ma, L.N. Robinson, H.C. Towle, ChREBP*Mlx is the principal mediator of glucose-induced gene expression in the liver, J. Biol. Chem. 281 (2006) 28721–28730. [15] B.M. Bolstad, R.A. Irizarry, M. Astrand, T.P. Speed, A comparison of normalization methods for high density oligonucleotide array data based on variance and bias, Bioinformatics 19 (2003) 185–193. [16] R.A. Irizarry, B.M. Bolstad, F. Collin, L.M. Cope, B. Hobbs, T.P. Speed, Summaries of Affymetrix GeneChip probe level data, Nucleic Acids Res. 31 (2003) e15. [17] R.A. Irizarry, B. Hobbs, F. Collin, Y.D. Beazer-Barclay, K.J. Antonellis, U. Scherf, et al., Exploration, normalization, and summaries of high density oligonucleotide array probe level data, Biostatistics 4 (2003) 249–264. [18] D. Croft, G. O’Kelly, G. Wu, R. Haw, M. Gillespie, L. Matthews, et al., Reactome: a database of reactions, pathways and biological processes, Nucleic Acids Res. 39 (2011) D691–7. [19] T. Ahn, E. Lee, N. Huh, T. Park, Personalized identification of altered pathways in cancer using accumulated normal tissue data, Bioinformatics 30 (2014) i422–9. [20] I. Diboun, L. Wernisch, C.A. Orengo, M. Koltzenburg, Microarray analysis after RNA amplification can detect pronounced differences in gene expression using limma, BMC Genomics 7 (2006) 252. [21] W. Cuiru, L. Ling, H. Yunxia, J. Ling, X. Ying, Z. Jinjuan, Effects of Bufei Capsule on pulmonary function and arterial blood gas of rat model of chronic obstructive pulmonary disease, Lishizhen Med. Mater. Med. Res. 23 (2012) 104–106. [22] Y. Benjamini, D. Drai, G. Elmer, N. Kafkafi, I. Golani, Controlling the false discovery rate in behavior genetics research, Behav. Brain Res. 125 (2001) 279–284. [23] A. Reiner, D. Yekutieli, Y. Benjamini, Identifying differentially expressed genes using false discovery rate controlling procedures, Bioinformatics 19 (2003) 368–375. [24] M.B. Eisen, P.T. Spellman, P.O. Brown, D. Botstein, Cluster analysis and display of genome-wide expression patterns, Proc. Natl. Acad. Sci. U. S. A. 95 (1998) 14863–14868. [25] A. Harel, A. Inger, G. Stelzer, L. Strichman-Almashanu, I. Dalah, M. Safran, et al., GIFtS: annotation landscape analysis with GeneCards, BMC Bioinf. 10 (2009) 348. [26] V. Baldin, J. Lukas, M.J. Marcote, M. Pagano, G. Draetta, Cyclin D1 is a nuclear protein required for cell cycle progression in G1, Genes. Dev. 7 (1993) 812–821. [27] R.M. Bohmer, E. Scharf, R.K. Assoian, Cytoskeletal integrity is required throughout the mitogen stimulation phase of the cell cycle and mediates the anchorage-dependent expression of cyclin D1, Mol. Biol. Cell 7 (1996) 101–111. [28] H. Akkiz, S. Bayram, A. Bekar, E. Akgollu, B. Ozdil, Cyclin D1 G870A polymorphism is associated with an increased risk of hepatocellular carcinoma in the Turkish population: case-control study, Cancer Epidemiol. 34 (2010) 298–302.
[29] M. Jin, S. Inoue, T. Umemura, J. Moriya, M. Arakawa, K. Nagashima, et al., Cyclin D1, p16 and retinoblastoma gene product expression as a predictor for prognosis in non-small cell lung cancer at stages I and II, Lung Cancer 34 (2001) 207–218. [30] A. Arnold, A. Papanikolaou, Cyclin D1 in breast cancer pathogenesis, J. Clin. Oncol. 23 (2005) 4215–4224. [31] K.E. Knudsen, J.A. Diehl, C.A. Haiman, E.S. Knudsen, Cyclin D1: polymorphism, aberrant splicing and cancer risk, Oncogene 25 (2006) 1620–1628. [32] J.W. Harper, G.R. Adami, N. Wei, K. Keyomarsi, S.J. Elledge, The p21 Cdk-interacting protein Cip1 is a potent inhibitor of G1 cyclin-dependent kinases, Cell 75 (1993) 805–816. [33] A.L. Gartel, S.K. Radhakrishnan, Lost in transcription: p21 repression, mechanisms, and consequences, Cancer Res. 65 (2005) 3980–3985. [34] R. Rodriguez, M. Meuth, Chk1 and p21 cooperate to prevent apoptosis during DNA replication fork stress, Mol. Biol. Cell 17 (2006) 402–412. [35] R. Klopfleisch, A.D. Gruber, Differential expression of cell cycle regulators p21, p27 and p53 in metastasizing canine mammary adenocarcinomas versus normal mammary glands, Res. Vet. Sci. 87 (2009) 91–96. [36] K. Ohta, H. Hoshino, J. Wang, S. Ono, Y. Iida, K. Hata, et al., MicroRNA-93 activates c-Met/PI3K/Akt pathway activity in hepatocellular carcinoma by directly inhibiting PTEN and CDKN1A, Oncotarget 6 (2015) 3211–3224. [37] Y. Xiong, H. Zhang, D. Beach, D type cyclins associate with multiple protein kinases and the DNA replication and repair factor PCNA, Cell 71 (1992) 505–514. [38] S.C. Wang, PCNA: a silent housekeeper or a potential therapeutic target, Trends Pharmacol. Sci. 35 (2014) 178–186. [39] J. Underhaug, O. Aubi, A. Martinez, Phenylalanine hydroxylase misfolding and pharmacological chaperones, Curr. Top. Med. Chem. 12 (2012) 2534–2545. [40] S.C. Correia, R. Resende, P.I. Moreira, C.M. Pereira, Alzheimer’s disease-related misfolded proteins and dysfunctional organelles on autophagy menu, DNA Cell Biol. 34 (2015) 261–273. [41] D.J. Selkoe, Folding proteins in fatal ways, Nature 426 (2003) 900–904. [42] J. Weinstein, F.W. Jacobsen, J. Hsu-Chen, T. Wu, L.G. Baum, A novel mammalian protein, p55CDC, present in dividing cells is associated with protein kinase activity and has homology to the Saccharomyces cerevisiae cell division cycle proteins Cdc20 and Cdc4, Mol. Cell. Biol. 14 (1994) 3350–3363. [43] J. Weinstein, Cell cycle-regulated expression, phosphorylation, and degradation of p55Cdc: a mammalian homolog of CDC20/Fizzy/slp1, J. Biol. Chem. 272 (1997) 28501–28511. [44] F. Jin, M. Hamada, L. Malureanu, K.B. Jeganathan, W. Zhou, D.E. Morbeck, et al., Cdc20 is critical for meiosis I and fertility of female mice, PLoS Genet. 6 (2010) e1001147. [45] H. Karra, H. Repo, I. Ahonen, E. Loyttyniemi, R. Pitkanen, M. Lintunen, et al., Cdc20 and securin overexpression predict short-term breast cancer survival, Br. J. Cancer 110 (2014) 2905–2913. [46] Z.Y. Ding, H.R. Wu, J.M. Zhang, G.R. Huang, D.D. Ji, Expression characteristics of CDC20 in gastric cancer and its correlation with poor prognosis, Int. J. Clin. Exp. Pathol. 7 (2014) 722–727. [47] Z. Wang, L. Wan, J. Zhong, H. Inuzuka, P. Liu, F.H. Sarkar, et al., Cdc20: a potential novel therapeutic target for cancer treatment, Curr. Pharm. Des. 19 (2013) 3210–3214. [48] J. Li, J.Z. Gao, J.L. Du, Z.X. Huang, L.X. Wei, Increased CDC20 expression is associated with development and progression of hepatocellular carcinoma, Int. J. Oncol. 45 (2014) 1547–1555.
Please cite this article in press as: H. Li, et al., Identification of personalized dysregulated pathways in hepatocellular carcinoma, Pathol. – Res. Pract (2017), http://dx.doi.org/10.1016/j.prp.2017.01.015
ARTICLE IN PRESS Pathology – Research and Practice xxx (2017) xxx–xxx
Contents lists available at ScienceDirect
Pathology – Research and Practice journal homepage: www.elsevier.com/locate/prp
Original article
Identification of personalized dysregulated pathways in hepatocellular carcinoma Hong Li 1 , Xiumei Jiang 1 , Shengjie Zhu, Lihong Sui ∗ Department of Oncology, Weihai Central Hospital, Weihai, 264400, Shandong, PR China
a r t i c l e
i n f o
Article history: Received 6 May 2016 Keywords: Hepatocellular carcinoma Differentially expressed genes Personalized dysregulated pathways Normal distribution analysis
a b s t r a c t Introduction: Hepatocellular carcinoma (HCC) is the most common liver malignancy, and ranks the fifth most prevalent malignant tumors worldwide. In general, HCC are detected until the disease is at an advanced stage and may miss the best chance for treatment. Thus, elucidating the molecular mechanisms is critical to clinical diagnosis and treatment for HCC. The purpose of this study was to identify dysregulated pathways of great potential functional relevance in the progression of HCC. Materials and methods: Microarray data of 72 pairs of tumor and matched non-tumor surrounding tissues of HCC were transformed to gene expression data. Differentially expressed genes (DEG) between patients and normal controls were identified using Linear Models for Microarray Analysis. Personalized dysregulated pathways were identified using individualized pathway aberrance score module. Results: 169 differentially expressed genes (DEG) were obtained with |logFC| ≥ 1.5 and P ≤ 0.01. 749 dysregulated pathways were obtained with P ≤ 0.01 in pathway statistics, and there were 93 DEG overlapped in the dysregulated pathways. After performing normal distribution analysis, 302 pathways with the aberrance probability ≥ 0.5 were identified. By ranking pathway with aberrance probability, the top 20 pathways were obtained. Only three DEGs (TUBA1C, TPR, CDC20) were involved in the top 20 pathways. Conclusion: These personalized dysregulated pathways and overlapped genes may give new insights into the underlying biological mechanisms in the progression of HCC. Particular attention can be focused on them for further research. © 2017 Elsevier GmbH. All rights reserved.
1. Introduction Hepatocellular carcinoma (HCC) is the most common liver malignancy, and ranks the fifth most prevalent malignant tumors worldwide [1]. Despite recent progress in anticancer, only 7% of patients with advanced HCC can expect to live for 5 years. In general, HCC are detected until the disease is at an advanced stage and may miss the best chance for treatment. Therefore, elucidating the molecular mechanisms is critical to clinical diagnosis and treatment for HCC. It is generally accepted that the occurrence of HCC is the result of genomic alteration [2,3]. In recent years, microarray technology has been used as an advanced high-throughput strategy for detecting gene expression profiles [4]. Tackels-Horne and colleagues successfully identified 842 over-regulated genes and 343 down-regulated
∗ Corresponding author at: Weihai Central Hospital, No.3, Mishan West Road, Wendeng District, Weihai 264400, Shandong, PR China. E-mail address: [email protected] (L. Sui). 1 These authors contributed equally.
genes involved in the pathogenesis of HCC [5]. Similarly, Chen and co-workers made a cross-study comparison for the tumor tissues and non-tumor tissues, and the results indicated that 1640 genes were significantly differentially expressed in the tumor tissues [6]. In addition, several growth factors, including EGFR, HDGF and IGF, have been proposed to be involved in the progression of HCC [7–9]. A previous study has demonstrated that FOXO-TXNIP pathway played a pivotal role in the inhibition of HCC growth by MK-801 [10]. Previous studies have proved that many critical genes and pathways are dysregulated during cancer initiation and progression [11]. Although there have been numerous studies on HCC pathogenesis, the results are not uniform and share only a small number of potential genes and pathways. The shortcoming highlights the importance of personalized pathway analysis. However, most current pathway analyses are mainly focused on discovering dysregulated pathways between two phenotype groups. Based on the shortcoming, a new approach is proposed to identify dysregulated pathways in an individual case. Our proposed method is based on the comparison of one cancer sample with many accumulated normal samples (we use “nRef” to refer to the accumulated normal
http://dx.doi.org/10.1016/j.prp.2017.01.015 0344-0338/© 2017 Elsevier GmbH. All rights reserved.
Please cite this article in press as: H. Li, et al., Identification of personalized dysregulated pathways in hepatocellular carcinoma, Pathol. – Res. Pract (2017), http://dx.doi.org/10.1016/j.prp.2017.01.015
G Model PRP-51730; No. of Pages 6
ARTICLE IN PRESS
2
H. Li et al. / Pathology – Research and Practice xxx (2017) xxx–xxx
samples). Personalized identification of dysregulated pathways is an important step toward understanding the disease mechanisms. An important clinical aspect of our method is providing a pathway interpretation of a single cancer, even though matched normal data are unavailable. Finally, individualized dysregulated pathways were identified by quantifying the aberrance of an individual sample’s pathway by comparing it with accumulated normal samples. In this study, differentially expressed genes (DEG) and individualized dysregulated pathways were identified. Microarray data of HCC patients were downloaded from ArrayExpress database [12]. The data were then preprocessed, and DEGs were identified using the LIMMA package. Subsequently, significantly dysregulated pathways between normal and cancer groups were identified. Finally, personalized dysregulated pathways were analyzed by using accumulated normal data. Among the 72 patients with HCC, the aberrance probability of each pathway was calculated. This study might provide useful information for exploring critical genes and individualized altered pathways, which can give insights into the diagnosis and treatment of HCC.
2.3. Pathway analysis Pathway analysis has become the first choice for capturing biologically and clinically relevant information. In our study, an individualized pathway aberrance score (iPAS) model was proposed to identify dysregulated pathways [21]. It makes special use of accumulated normal data to serve as the nRef. Data-processing procedures were described in more detail as follows. 2.3.1. Gene-level statistics The microarray data of normal samples were normalized using the quantile normalization method in preprocessCore package. For normal controls, the mean and standard deviations of the gene expression level were calculated. For individual tumor cases, quantile normalization [15] was performed after combining the single tumor microarray with all nRef samples. The gene-level statistics of individual tumor samples were standardized using the mean and standard deviation of the reference. The formula is shown as the following: zi =
2. Materials and methods 2.1. Datasets 2.1.1. Gene expression data and preprocessing Microarray expression data were downloaded from ArrayExpress database (http://www.ebi.ac.uk/arrayexpress/) using accession number E-GEOD-39791 [13]. The dataset included gene expression data from tumor and matched non-tumor surrounding tissues of 72 HCC patients who underwent surgical resection as the primary treatment. Gene expression data was generated using the HumanHT-12 Version 4.0 Expression Beadchip (Illumina) according to the annotation files. Linear Models for Microarray Data (LIMMA) was chosen to reprocess gene expression data. To eliminate influences of nonspecific hybridization, the background was corrected using robust multichip average (RMA) [14]. Normalization was performed with quantiles function. Then we used the MAS method to correct perfect match (PM)/mismatch (MM) [15], and the Medianpolish function was used to summarize expression data [16]. Then probes were mapped to gene symbols. The probe was discarded if it could not match any genes. The levels of probes were averaged as the final gene expression value if more than one probe was mapped to a single gene [17].
2.1.2. Pathway data All Homo sapiens-related biological pathway data were downloaded from Reactome database (http://www.reactome.org/) [18]. In general, pathways with a large number of genes are not easily understood by human experts [19]. Therefore, the pathways with a gene size >100 were filtered out. Meanwhile, the pathways without transformational gene symbols were removed. As a result, a total of 1022 pathways, including 4928 genes, were obtained for further analysis.
2.2. DEG analysis In the present study, Linear Models for Microarray Analysis (LIMMA) package of R were used to identify DEG between HCC patients and normal controls [20]. The values of |logFC| ≥ 1.5 and P ≤ 0.01 were selected as the cut-off criteria. In addition, the DEG belonging to pathways were ascertained by taking the intersection of identified DEG and all genes in pathways.
gTi − mean (gnRef ) stdev (gnRef )
(1)
where gTi represents the expression value of i-th gene belonging to the tumor cases, where mean (gnRef ) and stdev (gnRef ) represent the mean and standard deviation of the gene expression value of all reference samples, respectively. 2.3.2. Pathway-level statistics For each specific pathway, the standardized gene-level statistics of all genes were extracted. Then the gene-level statistics for all genes in a pathway were aggregated into a single pathway-level statistic. The model is as follows: iPAS =
zin
(2)
where zi symbolizes the standardized expression value of i-th gene belonging to the pathway, where n symbolizes the number of genes belonging to the pathway. 2.3.3. Dysregulated pathways Significance test was performed to assess dysregulated pathways associated with HCC. A generalized two-sample wilcoxon test was applied to assess the statistical significance of the pathwaylevel statistic. The false discovery rate (FDR) was used to correct the significance level [22,23]. The pathway with P < 0.01 was considered to be significantly dysregulated between normal and cancer groups. Moreover, a hierarchical clustering algorithm was applied to grouping the dysregulated pathways and the results were visualized with TreeView [24]. In addition, we obtained the common genes between known disease genes and dysregulated pathway genes. The known genes related to liver disease were selected, and a total of 39 genes were downloaded [25]. The pathways, including disease genes, were also ascertained. 2.3.4. Individualized dysregulated pathways Significance can be obtained against null distribution generated from normal samples on the basis of pathway-level statistics. The statistic of the null distribution was acquired via comparing every normal sample with all nRef samples. Then the p-value of each pathway was obtained after comparing a single cancer case with the null distribution generated above. A value with P < 0.05 was considered to indicate a statistically significant difference. Among the 72 patients with HCC, the aberrance probability of each significant pathway (P < 0.05) was calculated and ranked. The pathways with higher aberrance probability were considered to be more prone to alterations.
Please cite this article in press as: H. Li, et al., Identification of personalized dysregulated pathways in hepatocellular carcinoma, Pathol. – Res. Pract (2017), http://dx.doi.org/10.1016/j.prp.2017.01.015
G Model
ARTICLE IN PRESS
PRP-51730; No. of Pages 6
H. Li et al. / Pathology – Research and Practice xxx (2017) xxx–xxx
3
Table 1 Dysregulated pathways and the distribution of DEG. Term
Number of DEG
P-value
Probability
Phase 1 − Functionalization of compounds Phase II conjugation Cytochrome P450 − arranged by substrate type Xenobiotics Complement cascade Bile acid and bile salt metabolism CYP2E1 reactions Glucose metabolism Formation of Fibrin Clot (Clotting Cascade) Lipoprotein metabolism Regulation of Complement cascade Terminal pathway of complement
13 11 10 7 5 5 5 4 4 4 4 4
3.38E-10 1.55E-08 2.73E-10 7.49E-11 1.89E-10 3.48E-08 5.49E-11 6.62E-03 2.92E-10 6.11E-11 1.37E-08 3.62E-10
0.014 0.111 0.014 0 0.083 0.069 0 0.097 0.042 0 0.083 0.056
DEG denotes differential expressed genes.
Table 2 Known genes related to liver disease as well as shared in dysregulated pathways. No.
Genes
No.
Genes
No.
1 2 3 4 5 6 7 8 9 10 11 12
CCND1 CDKN1A PCNA IFNG IL6 ALB MET AXIN1 CASP3 SLC17A5 BCL2 JUN
13 14 15 16 17 18 19 20 21 22 23 24
CDH1 CTNNB1 PLAU FAS PTGS2 TP53 APC TGFB1 FASLG VEGFA TNF PIK3CA
25 26 27 28 29 30 31 32 33 34 35
3. Results 3.1. DEG analysis According to the criteria outlined (|logFC| ≥ 1.5, P ≤ 0.01), a total of 169 DEG were identified between tumor cases and normal controls, of which 17 were up-regulated and 152 were down-regulated. Among the DEG, 93 genes were as well as shared in pathways.
are shown in Table 3. Among them, degradation of the extracellular matrix and apoptotic cleavage of cellular proteins pathway contain 5 and 4 disease genes, respectively. 3.2.2. Individual dysregulated pathways Significance was obtained by null distribution calculated from the nRef. Statistical data from a single tumor sample were compared with the null distribution to yield p-value. Thus, the p-value of each pathway was obtained, and the pathway with p < 0.05 was screened. Subsequently, the aberrance probability of each significant pathway (p < 0.05) among the 72 patients was calculated and ranked. By ranking the pathways according to the aberrance probability, top 20 pathways are listed in Table 4. Based on our results, the “protein folding” pathway showed the highest aberrance probability (0.931). The other three pathways with the aberrance probability of 0.917 were also more prone to dysregulation, which were RNA Polymerase II Pre-transcription Events, Viral Messenger RNA Synthesis and Transport of Ribonucleoproteins into the Host Nucleus. Among the 72 patients, 933 pathways are able to dysregulate. A total of 302 pathways are more prone to alterations, the aberrance of which is more than 0.5. For further analysis, it was found that the top 20 ranked pathways were only involved in three DEG (TUBA1C, TPR, CDC20), and the other pathways do not contain DEG.
3.2. Pathway analysis 4. Discussion 3.2.1. Dysregulated pathways Wilcoxon test was performed to identify dysregulated pathways between normal controls and tumor cases. After FDR correction, a total of 749 pathways were significant (P < 0.01), of which only 165 pathways contain DEG and the remaining 584 pathways do not contain DEG. These pathways were ranked according to the number of DEG, and top 12 ranked pathways are listed in Table 1. It was also found that the aberrance probability of top ranked pathways was less than 0.12. For instance, the Phase 1 − Functionalization of compounds pathway, including 13 DEG, but the aberrance of which is only 0.014. Besides, a hierarchical clustering algorithm was applied to grouping the pathways and samples. The results for the 749 dysregulated pathways are shown in Fig. 1, and two features are evident. First, with a few exceptions, the clinical samples can be divided into two major clusters, one representing tumor samples, and the other representing normal samples. Second, pathways varied significantly among the tumor and normal samples. Furthermore, we obtained 39 disease genes, of which 35 genes (Table 2) were also shared in dysregulated pathways, such as CCND1, CDKN1A and PCNA. Meanwhile, the pathways containing disease genes were also ascertained. The results showed that there were 169 pathways including disease genes. The pathways were ranked by the number of disease genes, and top 9 ranked pathways
Cancers are generally caused by mutation of multiple genes or dysregulation of pathways. Identifying underlying genes and pathways would help to understand and diagnose cancers. The accumulation of microarray data makes it possible to detect critical genes and pathways in a more efficient way. This study aimed at analyzing underlying genes and personalized dysregulated pathways involved in HCC. LIMMA package was used to identify DEG between HCC patients and normal controls. The iPASbased approach was applied to explore individualized dysregulated pathways. In summary, many underlying genes and pathways were identified in our research. As a new approach for pathway analysis, iPAS was developed to identify altered pathways in an individual by making using of accumulated normal samples. Our proposed method captures biologically and clinically relevant information in a more accurate way compared to the interpretation of a single cancer sample in the context of cancer-only cohort. It is also useful in classifying unknown samples into cancer or normal groups. This study provides useful information to explore the dysregulated pathways and the underlying molecular mechanisms of HCC. Based on our results, a total of 169 genes were identified under the criteria of |logFC| ≥ 1.5 and P ≤ 0.01, of which 17 were up-regulated and 152 were down-regulated. Among the DEG,
Please cite this article in press as: H. Li, et al., Identification of personalized dysregulated pathways in hepatocellular carcinoma, Pathol. – Res. Pract (2017), http://dx.doi.org/10.1016/j.prp.2017.01.015
G Model PRP-51730; No. of Pages 6
ARTICLE IN PRESS
4
H. Li et al. / Pathology – Research and Practice xxx (2017) xxx–xxx
Fig 1. A hierarchical cluster for the 749 deregulated pathways. Rows represent individual pathways and columns represent individual tissue samples. The color reflects different samples, red representing normal samples, and blue representing tumor samples.
Table 3 Dysregulated pathways and the distribution of disease genes. Term
Disease genes
Probability
Degradation of the extracellular matrix Apoptotic cleavage of cellular proteins Degradation of beta-catenin by the destruction complex Transcriptional regulation of white adipocyte differentiation Cyclin D associated events in G1 G1 Phase Intrinsic Pathway for Apoptosis Death Receptor Signalling Beta-catenin phosphorylation cascade
CASP3,CDH1,MMP2,MMP9,TIMP1 APC,CASP3,CDH1,CTNNB1 APC,AXIN1,CTNNB1 PPARG,TGFB1,TNF CCND1,CDKN1A,CDKN2A CCND1,CDKN1A,CDKN2A BCL2,CASP3,TP53 FAS,FASLG,TNF APC,AXIN1,CTNNB1
0.292 0.292 0.861 0.708 0.806 0.806 0.806 0.306 0.375
Table 4 The top 20 ranked pathways and distribution of DEG. Pathways
No.
Probability
DEG
Protein folding RNA Polymerase II Pre-transcription Events Viral Messenger RNA Synthesis Transport of Ribonucleoproteins into the Host Nucleus Transcription of the HIV genome HIV Transcription Initiation RNA Polymerase II HIV Promoter Escape RNA Polymerase II Promoter Escape RNA Polymerase II Transcription Initiation RNA Polymerase II Transcription Initiation And Promoter Clearance RNA PolymeraseII Transcription Pre-Initiation And Promoter Opening Transport of the SLBP Dependant Mature mRNA Transport of the SLBP independent Mature mRNA APC/C:Cdc20 mediated degradation of Cyclin B Translesion Synthesis by POLH APC/C-mediated degradation of cell cycle proteins Regulation of mitotic cell cycle
67 66 66 66 65 65 65 65 65 65 65 65 65 65 65 64 64
0.931 0.917 0.917 0.917 0.903 0.903 0.903 0.903 0.903 0.903 0.903 0.903 0.903 0.903 0.903 0.889 0.889
TUBA1C – TPR TPR – – – – – – – TPR TPR CDC20 – CDC20 CDC20
93 were also shared in pathways. Subsequently, pathway analysis was carried out to assess potential pathways associated with HCC. Wilcoxon test was applied to identify significantly dysregulated pathways between normal controls and tumor cases. The results indicated that 749 pathways with p < 0.01 were significantly altered. Among these pathways, only 165 pathways contain DEG.
It was found that the number of DEG is inconsistent with statistical significance. It is possible that genes do not work in isolation but interact with each other to perform their functions. Therefore, dysregulated pathways can serve as better biomarkers compared with single genes.
Please cite this article in press as: H. Li, et al., Identification of personalized dysregulated pathways in hepatocellular carcinoma, Pathol. – Res. Pract (2017), http://dx.doi.org/10.1016/j.prp.2017.01.015
G Model PRP-51730; No. of Pages 6
ARTICLE IN PRESS H. Li et al. / Pathology – Research and Practice xxx (2017) xxx–xxx
In recent years, many genes have been identified to be related to liver disease. In this article, 39 disease genes were downloaded, of which 35 genes were also shared in dysregulated pathways. These genes once again were validated, such as CCND1, CDKN1A and PCNA. Meanwhile, the pathways containing disease genes were also ascertained, and only 169 dysregulated pathways include disease genes, such as degradation of the extracellular matrix and apoptotic cleavage of cellular proteins pathway. This indicates that some key genes cannot be discovered, and more attention can be paid to them. These potential genes and pathways may give insights into disease mechanisms and even drug targets. CCND1 is a protein coding gene and encodes the cyclin-D1 protein, which is responsible for cell cycle progression [26]. In G1 phase, cyclin-D1 is synthesized rapidly by the inducing action of mitogenic growth factors and accumulated in the nucleus[27]. Experimental study revealed that Cyclin-D1 may play an important role in HCC [28]. In addition, amplification and over-expression of this gene play pivotal roles in the development of various tumors, such as non-small cell lung cancer [29], breast cancer [30], and prostate cancer[31]. CDKN1A (P21) is a cyclin-dependent kinase inhibitor, and can regulate cell cycle progression and cellular senescence [32,33]. The gene is an important intermediate by which p53 mediates its role as an inhibitor of cellular proliferation in response to DNA damage [34]. A previous study indicated that P21 was upregulated in metastatic canine mammary tumor [35]. P21 plays an important role during HCC progression [36]. PCNA is closely related to cell DNA synthesis and can be used as evaluation index for the cell proliferation state. The protein can interact with CDKN1A and plays a regulatory role in DNA replication and DNA damage repair [37]. The research investigating the relationship between PCNA and tumor has been a hot field, and PCNA is a potential therapeutic target in cancer therapy [38]. These known disease genes, which were also included in dysregulated pathways, can be considered as critical therapeutic target. Pathway-based identification of cancer has been proposed to be a robust method, especially individualized pathway analysis provides a more effective way in cancer identification. Null distribution was applied to yield the P −value of each pathway. The pathway with P < 0.05 was screened among the 72 tumor samples. Furthermore, the significant aberrance probability of each pathway was calculated and ranked. The results indicated that protein folding pathway had the highest aberrance probability. Protein folding pathway consists of 52 genes, of which TUBA1C is differentially expressed. Protein folding is an essential process in which polypeptide can self-assemble into its characteristic and functional three-dimensional structure. The correct three-dimensional structure is crucial to maintaining the protein function. Recent evidence suggests that misfolded protein can cause a variety of diseases, such as Phenylketonuria, Alzheimer and Parkinson’s diseases [39–41]. Pathways that are commonly dysregulated across the overwhelming majority of cancer patients may be critical in identifying cancer. These identified pathways were considered closely related to HCC. For further analysis, it was revealed that top 20 ranked pathways are only involved in three DEG (TUBA1C, TPR, CDC20). This is not consistent with traditional methods that detect DEG between tumor cases and normal controls. The power of pathway-level statistics not only depends on the proportion of DEG in a pathway, but is also involved in the size of the pathway and the amount of correlation between genes in the pathway. It is believed that dysregulated pathways show greater advantage than single genes in identifying cancer. CDC20 encodes cell-division cycle protein 20, which is a regulatory protein of cell division [42,43]. Its most important function is to activate the anaphase-promoting complex (APC/C), and plays an important role in tumorigenesis and progression of multiple tumors [44]. Accumulating evidence indicated that CDC20 may function as an oncogene and was over-expressed in various types
5
of human tumors, such as breast cancer, ovarian cancer and gastric cancer [45–47]. Increased expression of CDC20 has also been demonstrated to be associated with the development and progression of HCC [48]. However, TPR and TUBA1C show poor research on disease mechanisms. The three DEG (TUBA1C, TPR, CDC20), which also belong to the top ranked pathways, are considered pivotal in the progression of HCC. More attention should be paid to them, and they may be regarded as a promising therapeutic target for HCC. In this study, many underlying genes and personalized dysregulated pathways were identified. Among them, a part of them have been reported to be associated with HCC, and the others showed poor research. The results need to be validated in a prospective study, and all of these would become another perspective research. 5. Conclusions In this study, underlying genes and personalized dysregulated pathways were identified. First, LIMMA package was used to identify DEG and 169 genes were significantly differentially expressed. Next, the Wilcoxon test was performed to identify dysregulated pathways between normal and tumor groups. The analysis revealed that 749 pathways were significantly dysregulated between normal and cancer groups. Meanwhile, disease genes related to liver were also obtained. A total of 39 disease genes were downloaded, of which 35 were included in dysregulated pathways. Dysregulated pathways, in particular those pathways including disease genes, might provide insights into disease mechanisms and drug targets. Finally, an iPAS-based approach was employed to identify personalized dysregulated pathways. Individualized pathway analysis indicated that protein folding pathway had the highest aberrance probability. Furthermore, it was also revealed that the top ranked pathways were only involved in three DEG: TPR, TUBA1C and CDC20. Dysregulated pathways act as more effective biomarkers in identifying cancers. These personalized dysregulated pathways and overlap genes may give new insights into the underlying biological mechanisms driving the progression of HCC. Conflict of interest None. References [1] A. Jemal, F. Bray, M.M. Center, J. Ferlay, E. Ward, D. Forman, Global cancer statistics, CA. Cancer J. Clin. 61 (2011) 69–90. [2] X.Y. Guan, Y. Fang, J. Sham, D. Kwong, Y. Zhang, Q. Liang, et al., Recurrent chromosome alterations in hepatocellular carcinoma detected by comparative genomic hybridization, Genes Chromosomes Cancer 30 (2001) 110. [3] S.S. Thorgeirsson, J.W. Grisham, Molecular pathogenesis of human hepatocellular carcinoma, Nat. Genet. 31 (2002) 339–346. [4] A. Suriawinata, R. Xu, An update on the molecular genetics of hepatocellular carcinoma, Semin. Liver Dis. 24 (2004) 77–88. [5] D. Tackels-Horne, M.D. Goodman, A.J. Williams, D.J. Wilson, T. Eskandari, L.M. Vogt, et al., Identification of differentially expressed genes in hepatocellular carcinoma and metastatic liver tumors by oligonucleotide expression profiling, Cancer 92 (2001) 395–405. [6] X. Chen, S.T. Cheung, S. So, S.T. Fan, C. Barry, J. Higgins, et al., Gene expression patterns in human liver cancers, Mol. Biol. Cell 13 (2002) 1929–1939. [7] M.B. Thomas, D. Jaffe, M.M. Choti, J. Belghiti, S. Curley, Y. Fong, et al., Hepatocellular carcinoma: consensus recommendations of the national cancer institute clinical trials planning meeting, J. Clin. Oncol. 28 (2010) 3994–4005. [8] H. Enomoto, H. Nakamura, W. Liu, S. Nishiguchi, Hepatoma-Derived growth factor: its possible involvement in the progression of hepatocellular carcinoma, Int. J. Mol. Sci. 16 (2015) 14086–14097. [9] M. Enguita-German, P. Fortes, Targeting the insulin-like growth factor pathway in hepatocellular carcinoma, World J. Hepatol. 6 (2014) 716–737. [10] F. Yamaguchi, Y. Hirata, H. Akram, K. Kamitori, Y. Dong, L. Sui, et al., FOXO/TXNIP pathway is involved in the suppression of hepatocellular carcinoma growth by glutamate antagonist MK-801, BMC Cancer 13 (2013) 468.
Please cite this article in press as: H. Li, et al., Identification of personalized dysregulated pathways in hepatocellular carcinoma, Pathol. – Res. Pract (2017), http://dx.doi.org/10.1016/j.prp.2017.01.015
G Model PRP-51730; No. of Pages 6
ARTICLE IN PRESS
6
H. Li et al. / Pathology – Research and Practice xxx (2017) xxx–xxx
[11] A.H. Bild, G. Yao, J.T. Chang, Q. Wang, A. Potti, D. Chasse, et al., Oncogenic pathway signatures in human cancers as a guide to targeted therapies, Nature 439 (2006) 353–357. [12] H. Parkinson, M. Kapushesky, M. Shojatalab, N. Abeygunawardena, R. Coulson, A. Farne, et al., ArrayExpress–a public database of microarray experiments and gene expression profiles, Nucleic Acids Res. 35 (2007) D747–D750. [13] J.H. Kim, B.H. Sohn, H.S. Lee, S.B. Kim, J.E. Yoo, Y.Y. Park, et al., Genomic predictors for recurrence patterns of hepatocellular carcinoma: model derivation and validation, PLoS Med. 11 (2014) e1001770. [14] L. Ma, L.N. Robinson, H.C. Towle, ChREBP*Mlx is the principal mediator of glucose-induced gene expression in the liver, J. Biol. Chem. 281 (2006) 28721–28730. [15] B.M. Bolstad, R.A. Irizarry, M. Astrand, T.P. Speed, A comparison of normalization methods for high density oligonucleotide array data based on variance and bias, Bioinformatics 19 (2003) 185–193. [16] R.A. Irizarry, B.M. Bolstad, F. Collin, L.M. Cope, B. Hobbs, T.P. Speed, Summaries of Affymetrix GeneChip probe level data, Nucleic Acids Res. 31 (2003) e15. [17] R.A. Irizarry, B. Hobbs, F. Collin, Y.D. Beazer-Barclay, K.J. Antonellis, U. Scherf, et al., Exploration, normalization, and summaries of high density oligonucleotide array probe level data, Biostatistics 4 (2003) 249–264. [18] D. Croft, G. O’Kelly, G. Wu, R. Haw, M. Gillespie, L. Matthews, et al., Reactome: a database of reactions, pathways and biological processes, Nucleic Acids Res. 39 (2011) D691–7. [19] T. Ahn, E. Lee, N. Huh, T. Park, Personalized identification of altered pathways in cancer using accumulated normal tissue data, Bioinformatics 30 (2014) i422–9. [20] I. Diboun, L. Wernisch, C.A. Orengo, M. Koltzenburg, Microarray analysis after RNA amplification can detect pronounced differences in gene expression using limma, BMC Genomics 7 (2006) 252. [21] W. Cuiru, L. Ling, H. Yunxia, J. Ling, X. Ying, Z. Jinjuan, Effects of Bufei Capsule on pulmonary function and arterial blood gas of rat model of chronic obstructive pulmonary disease, Lishizhen Med. Mater. Med. Res. 23 (2012) 104–106. [22] Y. Benjamini, D. Drai, G. Elmer, N. Kafkafi, I. Golani, Controlling the false discovery rate in behavior genetics research, Behav. Brain Res. 125 (2001) 279–284. [23] A. Reiner, D. Yekutieli, Y. Benjamini, Identifying differentially expressed genes using false discovery rate controlling procedures, Bioinformatics 19 (2003) 368–375. [24] M.B. Eisen, P.T. Spellman, P.O. Brown, D. Botstein, Cluster analysis and display of genome-wide expression patterns, Proc. Natl. Acad. Sci. U. S. A. 95 (1998) 14863–14868. [25] A. Harel, A. Inger, G. Stelzer, L. Strichman-Almashanu, I. Dalah, M. Safran, et al., GIFtS: annotation landscape analysis with GeneCards, BMC Bioinf. 10 (2009) 348. [26] V. Baldin, J. Lukas, M.J. Marcote, M. Pagano, G. Draetta, Cyclin D1 is a nuclear protein required for cell cycle progression in G1, Genes. Dev. 7 (1993) 812–821. [27] R.M. Bohmer, E. Scharf, R.K. Assoian, Cytoskeletal integrity is required throughout the mitogen stimulation phase of the cell cycle and mediates the anchorage-dependent expression of cyclin D1, Mol. Biol. Cell 7 (1996) 101–111. [28] H. Akkiz, S. Bayram, A. Bekar, E. Akgollu, B. Ozdil, Cyclin D1 G870A polymorphism is associated with an increased risk of hepatocellular carcinoma in the Turkish population: case-control study, Cancer Epidemiol. 34 (2010) 298–302.
[29] M. Jin, S. Inoue, T. Umemura, J. Moriya, M. Arakawa, K. Nagashima, et al., Cyclin D1, p16 and retinoblastoma gene product expression as a predictor for prognosis in non-small cell lung cancer at stages I and II, Lung Cancer 34 (2001) 207–218. [30] A. Arnold, A. Papanikolaou, Cyclin D1 in breast cancer pathogenesis, J. Clin. Oncol. 23 (2005) 4215–4224. [31] K.E. Knudsen, J.A. Diehl, C.A. Haiman, E.S. Knudsen, Cyclin D1: polymorphism, aberrant splicing and cancer risk, Oncogene 25 (2006) 1620–1628. [32] J.W. Harper, G.R. Adami, N. Wei, K. Keyomarsi, S.J. Elledge, The p21 Cdk-interacting protein Cip1 is a potent inhibitor of G1 cyclin-dependent kinases, Cell 75 (1993) 805–816. [33] A.L. Gartel, S.K. Radhakrishnan, Lost in transcription: p21 repression, mechanisms, and consequences, Cancer Res. 65 (2005) 3980–3985. [34] R. Rodriguez, M. Meuth, Chk1 and p21 cooperate to prevent apoptosis during DNA replication fork stress, Mol. Biol. Cell 17 (2006) 402–412. [35] R. Klopfleisch, A.D. Gruber, Differential expression of cell cycle regulators p21, p27 and p53 in metastasizing canine mammary adenocarcinomas versus normal mammary glands, Res. Vet. Sci. 87 (2009) 91–96. [36] K. Ohta, H. Hoshino, J. Wang, S. Ono, Y. Iida, K. Hata, et al., MicroRNA-93 activates c-Met/PI3K/Akt pathway activity in hepatocellular carcinoma by directly inhibiting PTEN and CDKN1A, Oncotarget 6 (2015) 3211–3224. [37] Y. Xiong, H. Zhang, D. Beach, D type cyclins associate with multiple protein kinases and the DNA replication and repair factor PCNA, Cell 71 (1992) 505–514. [38] S.C. Wang, PCNA: a silent housekeeper or a potential therapeutic target, Trends Pharmacol. Sci. 35 (2014) 178–186. [39] J. Underhaug, O. Aubi, A. Martinez, Phenylalanine hydroxylase misfolding and pharmacological chaperones, Curr. Top. Med. Chem. 12 (2012) 2534–2545. [40] S.C. Correia, R. Resende, P.I. Moreira, C.M. Pereira, Alzheimer’s disease-related misfolded proteins and dysfunctional organelles on autophagy menu, DNA Cell Biol. 34 (2015) 261–273. [41] D.J. Selkoe, Folding proteins in fatal ways, Nature 426 (2003) 900–904. [42] J. Weinstein, F.W. Jacobsen, J. Hsu-Chen, T. Wu, L.G. Baum, A novel mammalian protein, p55CDC, present in dividing cells is associated with protein kinase activity and has homology to the Saccharomyces cerevisiae cell division cycle proteins Cdc20 and Cdc4, Mol. Cell. Biol. 14 (1994) 3350–3363. [43] J. Weinstein, Cell cycle-regulated expression, phosphorylation, and degradation of p55Cdc: a mammalian homolog of CDC20/Fizzy/slp1, J. Biol. Chem. 272 (1997) 28501–28511. [44] F. Jin, M. Hamada, L. Malureanu, K.B. Jeganathan, W. Zhou, D.E. Morbeck, et al., Cdc20 is critical for meiosis I and fertility of female mice, PLoS Genet. 6 (2010) e1001147. [45] H. Karra, H. Repo, I. Ahonen, E. Loyttyniemi, R. Pitkanen, M. Lintunen, et al., Cdc20 and securin overexpression predict short-term breast cancer survival, Br. J. Cancer 110 (2014) 2905–2913. [46] Z.Y. Ding, H.R. Wu, J.M. Zhang, G.R. Huang, D.D. Ji, Expression characteristics of CDC20 in gastric cancer and its correlation with poor prognosis, Int. J. Clin. Exp. Pathol. 7 (2014) 722–727. [47] Z. Wang, L. Wan, J. Zhong, H. Inuzuka, P. Liu, F.H. Sarkar, et al., Cdc20: a potential novel therapeutic target for cancer treatment, Curr. Pharm. Des. 19 (2013) 3210–3214. [48] J. Li, J.Z. Gao, J.L. Du, Z.X. Huang, L.X. Wei, Increased CDC20 expression is associated with development and progression of hepatocellular carcinoma, Int. J. Oncol. 45 (2014) 1547–1555.
Please cite this article in press as: H. Li, et al., Identification of personalized dysregulated pathways in hepatocellular carcinoma, Pathol. – Res. Pract (2017), http://dx.doi.org/10.1016/j.prp.2017.01.015